HVM domains (Hardware VM), in contrast to PV domains (Paravirtualized domains), allow one to create domains based on any OS for which one has an installation ISO. For example, this allows one to have Windows-based VMs in Qubes.
Interested readers might want to check [this article](https://blog.invisiblethings.org/2012/03/03/windows-support-coming-to-qubes.html) to learn why it took so long for Qubes OS to support HVM domains (Qubes 1 only supported Linux based PV domains). As of Qubes 4, every VM is PVH by default, except those with attached PCI devices which are HVM. [See here](https://blog.invisiblethings.org/2017/07/31/qubes-40-rc1.html) for a discussion of the switch to HVM from R3.2's PV, and [here](/news/2018/01/11/qsb-37/) for changing the default to PVH.
With a GUI: in Qubes Manager VM creation dialog box choose the "Standalone qube not based on a template" type. If "install system from device" is selected (which is by default), then `virt_mode` will be set to `hvm` automatically. Otherwise, open the newly created VM's Qube Settings GUI and in the "Advanced" tab select "HVM" in the virtualization mode drop-down list.
You will have to boot the VM with the installation media "attached" to it. You may either use the GUI or command line instructions. That can be accomplished in three ways:
For security reasons you should *never* copy untrusted data to dom0. Qubes doesn't provide any easy to use mechanism for copying files between VMs and Dom0 anyway and generally tries to discourage such actions.
Next, the VM will start booting from the attached installation media. Depending on the OS being installed in the VM, one might be required to start the VM several times (as is the case with Windows 7 installations) because whenever the installer wants to "reboot the system" it actually shuts down the VM and Qubes won't automatically start it. Several invocations of `qvm-start` command (as shown above) might be needed.
Just like standard paravirtualized AppVMs, the HVM domains get fixed IP addresses centrally assigned by Qubes. Normally Qubes agent scripts (or services on Windows) running within each AppVM are responsible for setting up networking within the VM according the configuration created by Qubes (through [keys](/doc/vm-interface/#qubesdb) exposed by dom0 to the VM). Such centrally managed networking infrastructure allows for [advanced networking configuration](https://blog.invisiblethings.org/2011/09/28/playing-with-qubes-networking-for-fun.html).
A generic HVM domain such as a standard Windows or Ubuntu installation, however, has no Qubes agent scripts running inside it initially and thus requires manual networking configuration so that it match the values assigned by Qubes for this domain.
Even though we do have a small DHCP server that runs inside HVM untrusted stub domain to make the manual network configuration not necessary for many VMs, this won't work for most modern Linux distributions which contain Xen networking PV drivers (but not Qubes tools) built in which bypass the stub-domain networking (their net frontends connect directly to the net backend in the netvm). In this instance our DHCP server is not useful.
In order to manually configure networking in a VM, one should first find out the IP/netmask/gateway assigned to the particular VM by Qubes. This can be seen e.g. in the Qubes Manager in the VM's properties:
Alternatively, one can use the `qvm-ls -n` command to obtain the same information and configure the networking within the HVM according to those settings (IP/netmask/gateway).
Just like normal AppVMs, the HVM domains can also be cloned either using a command-line `qvm-clone` command or via manager's 'Clone VM' option in the right-click menu.
The cloned VM will get identical root and private images and will essentially be identical to the original VM except that it will get a different MAC address for the networking interface:
Note how the MAC addresses differ between those two otherwise identical VMs. The IP addresses assigned by Qubes will also be different of course to allow networking to function properly:
HVM domains (including Windows VMs) can be [assigned PCI devices](/doc/assigning-devices/) just like normal AppVMs. E.g. one can assign one of the USB controllers to the Windows VM and should be able to use various devices that require Windows software, such as phones, electronic devices that are configured via FTDI, etc.
One problem at the moment however, is that after the whole system gets suspended into S3 sleep and subsequently resumed, some attached devices may stop working and should be restarted within the VM. This can be achieved under a Windows HVM by opening the Device Manager, selecting the actual device (such as a USB controller), 'Disabling' the device, and then 'Enabling' the device again. This is illustrated on the screenshot below:
